Choose Create or Explore
The + and / controls reveal supported actions. Explore helps prioritize work, prepare interview answers, and draft text from bounded Workspace and Opportunity context.
Full-stack web product · Job-search operating system
Roleway
Roleway turns a scattered job search into focused Workspaces and clear next steps. Keep roles, interviews, people, and documents together, then ask its AI Agent to explore your search or prepare a Workspace, task, note, or Next Action for approval.
AI Agent
Explore your search, turn a request into a proposal, and stay in control of every saved change.
Fill the gaps through chat
Agent asks for missing details one at a time and clarifies the target Opportunity, Workspace, or date. It can propose a Workspace, task, note, or Next Action.
Approve the exact change
Review the destination and proposed fields before approval. Applied means the record was saved; Open takes you to the result. Expired or stale Next Action proposals require a fresh review.
Product showcase
Inside Roleway
Selected product surfaces and public demonstrations. Demo images use illustrative records; captions identify those examples.
Problem and response
The problem
A serious job search spreads across browser tabs, spreadsheets, calendars, documents, and notes. Different career targets blur together, follow-ups are missed, preparation loses context, and it becomes difficult to decide what deserves attention next.
The approach
An Account owns a Career Profile and multiple Workspaces, each with its own Jobs, Opportunities, preferences, people, documents, goals, and results. Home orders current work and each Opportunity retains its dossier and one Next Action. Agent helps users explore selected search context and create through conversation: clarify the destination, gather missing details, review the exact proposal, and open the saved result after approval.
The workflow
Capture, then commit
Save a listing to Inbox. Track it as an Opportunity when it deserves active work, with a stage and one concrete Next Action.
Prepare with the full record
Keep interviews, contacts, preparation tasks, and document versions connected to the right Opportunity and Workspace.
Know what needs attention
Home brings due work and upcoming interviews together. Notifications and Insights reflect activity recorded in the product.
Keep AI optional
Connect your own provider for Agent; core tracking works without it. Provider charges are separate from Roleway capacity limits.
Architecture
Roleway is a pnpm and Turborepo TypeScript monorepo. A Next.js application composes shared workflow rules and Zod schemas, while Supabase provides authentication, PostgreSQL, row-level security, and the durable records behind Workspaces and Agent runs.
Next.js application
Server Components load authenticated Workspace data and Server Actions handle mutations. Client components stay focused on interactive controls, editors, collection views, onboarding, and the installable PWA shell.
Core and schema packages
Opportunity transitions, workflow labels, Agent permissions, and Zod boundary schemas live in shared packages so domain rules can be tested independently of route components.
Owner-scoped PostgreSQL
Supabase migrations are the source of truth for Workspaces, Jobs, Opportunities, tasks, contacts, interviews, documents, notifications, analytics, and Agent records. Row-level security and server checks enforce ownership.
Provider-neutral Agent boundary
Next.js Server Actions gather bounded account context before one generation request through custom HTTP adapters for OpenAI, Anthropic, Gemini, OpenRouter, or compatible endpoints. Zod validates the structured answer and proposals before storage. Responses appear after completion; this implementation does not use Vercel AI SDK or token streaming. User-supplied credentials are encrypted with AES-256-GCM.
Database-authorized Agent actions
Agent proposes Workspaces, Opportunity tasks, Next Actions, and notes. Explicit approval invokes PostgreSQL functions that recheck ownership, proposal status, expiry, and domain rules in a transaction. Stale Next Action proposals are rejected instead of overwriting intervening edits. Explore uses bounded snapshots, not exhaustive account search. Documents contribute metadata; their full bodies, notes, activity history, and Help Center articles are not retrieved into model context.
Key product and engineering decisions
These decisions affect what users can do, where product rules live, and how the system handles failure or change.
Give each search its own Workspace
An Account can hold separate Workspaces for different career targets. Preferences, Jobs, Opportunities, documents, people, goals, notifications, and analytics stay isolated instead of becoming one shared junk drawer.
Separate discovery from committed work
A Job stores discovered listing data. Tracking it creates an Opportunity with a stage and one concrete Next Action, so the active list represents deliberate work rather than another bookmarks folder.
Prepare with Agent, approve every change
Core tracking works without AI. Create and Explore are discoverable from the same + and / menu. Agent asks for missing details in chat, then presents the exact destination, text, and date for approval. Applications and employer messages remain user actions.
Protect data at two boundaries
PostgreSQL row-level security scopes records to their owner and Workspace, while server actions independently authenticate callers and constrain mutations by owner and Workspace IDs.
What you can inspect
The live product and public repository provide concrete evidence of the implemented features, architecture, tests, and documentation.
- The deployed product includes multiple Workspaces, Home, Job Inbox, list and board Opportunity views, detailed Opportunity dossiers, contacts, interviews, documents, entity search, notifications, Insights, settings, export, account deletion, and an admin console.
- Public job URLs are fetched server-side with private-network blocking, redirect and size limits, and a per-user quota; Ashby, Greenhouse, and Lever use public job-board endpoints with JSON-LD and metadata fallbacks.
- The public Agent showcase demonstrates next-step exploration and conversational Workspace/task creation with illustrative data. The live creation guide documents all four supported changes, correction, rejection, and recovery. These demos do not execute live AI calls.
- Unit tests cover workflow transitions, provider parsing, encrypted credentials, rich text, dates, URL safety, and validation. Playwright covers signup, onboarding, Workspaces, capture, Opportunity progression, application records, documents, interviews, Home, search, export, authorization, and account deletion.
- The Apache-2.0 repository is self-hostable. The PWA caches versioned static assets while authenticated pages and API responses remain network-only.